Cycles: Refactor Image Texture limits.

Instead of treating Fermi GPU limits as default,
and overriding them for other devices,
we now nicely set them for each platform.

* Due to setting values for all platforms,
we don't have to offset the slot id for OpenCL anymore,
as the image manager wont add float images for OpenCL now.

* Bugfix: TEX_NUM_FLOAT_IMAGES was always 5, even for CPU,
so the code in svm_image.h clamped float textures with alpha on CPU after the 5th slot.

Reviewers: #cycles, brecht

Reviewed By: #cycles, brecht

Subscribers: brecht

Differential Revision: https://developer.blender.org/D1925

intern/cycles/kernel/svm/svm_image.h

@@ -16,6 +16,15 @@
 
 CCL_NAMESPACE_BEGIN
 
+/* Float textures on various devices. */
+#if defined(__KERNEL_CPU__)
+  #define TEX_NUM_FLOAT_IMAGES	TEX_NUM_FLOAT_IMAGES_CPU
+#elif defined(__KERNEL_CUDA__)
+  #define TEX_NUM_FLOAT_IMAGES	TEX_NUM_FLOAT_IMAGES_CUDA
+#else
+  #define TEX_NUM_FLOAT_IMAGES	TEX_NUM_FLOAT_IMAGES_OPENCL
+#endif
+
 #ifdef __KERNEL_OPENCL__
 
 /* For OpenCL all images are packed in a single array, and we do manual lookup
@@ -50,12 +59,6 @@ ccl_device_inline float svm_image_texture_frac(float x, int *ix)
 
 ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, uint srgb, uint use_alpha)
 {
-	/* first slots are used by float textures, which are not supported here */
-	if(id < TEX_NUM_FLOAT_IMAGES)
-		return make_float4(1.0f, 0.0f, 1.0f, 1.0f);
-
-	id -= TEX_NUM_FLOAT_IMAGES;
-
 	uint4 info = kernel_tex_fetch(__tex_image_packed_info, id);
 	uint width = info.x;
 	uint height = info.y;

intern/cycles/render/image.cpp

@@ -30,16 +30,46 @@
 
 CCL_NAMESPACE_BEGIN
 
-ImageManager::ImageManager()
+ImageManager::ImageManager(const DeviceInfo& info)
 {
 	need_update = true;
 	pack_images = false;
 	osl_texture_system = NULL;
 	animation_frame = 0;
 
-	tex_num_images = TEX_NUM_IMAGES;
+	/* Set image limits */
-	tex_num_float_images = TEX_NUM_FLOAT_IMAGES;
+
-	tex_image_byte_start = TEX_IMAGE_BYTE_START;
+	/* CPU */
+	if(info.type == DEVICE_CPU) {
+		tex_num_byte_images = TEX_NUM_BYTE_IMAGES_CPU;
+		tex_num_float_images = TEX_NUM_FLOAT_IMAGES_CPU;
+		tex_image_byte_start = TEX_IMAGE_BYTE_START_CPU;
+	}
+	/* CUDA (Fermi) */
+	else if((info.type == DEVICE_CUDA || info.type == DEVICE_MULTI) && !info.extended_images) {
+		tex_num_byte_images = TEX_NUM_BYTE_IMAGES_CUDA;
+		tex_num_float_images = TEX_NUM_FLOAT_IMAGES_CUDA;
+		tex_image_byte_start = TEX_IMAGE_BYTE_START_CUDA;
+	}
+	/* CUDA (Kepler and above) */
+	else if((info.type == DEVICE_CUDA || info.type == DEVICE_MULTI) && info.extended_images) {
+		tex_num_byte_images = TEX_NUM_BYTE_IMAGES_CUDA_KEPLER;
+		tex_num_float_images = TEX_NUM_FLOAT_IMAGES_CUDA_KEPLER;
+		tex_image_byte_start = TEX_IMAGE_BYTE_START_CUDA_KELPER;
+	}
+	/* OpenCL */
+	else if(info.pack_images) {
+		tex_num_byte_images = TEX_NUM_BYTE_IMAGES_OPENCL;
+		tex_num_float_images = TEX_NUM_FLOAT_IMAGES_OPENCL;
+		tex_image_byte_start = TEX_IMAGE_BYTE_START_OPENCL;
+	}
+	/* Should never happen */
+	else {
+		tex_num_byte_images = 0;
+		tex_num_float_images = 0;
+		tex_image_byte_start = 0;
+		assert(0);
+	}
 }
 
 ImageManager::~ImageManager()
@@ -60,21 +90,6 @@ void ImageManager::set_osl_texture_system(void *texture_system)
 	osl_texture_system = texture_system;
 }
 
-void ImageManager::set_extended_image_limits(const DeviceInfo& info)
-{
-	if(info.type == DEVICE_CPU) {
-		tex_num_images = TEX_EXTENDED_NUM_IMAGES_CPU;
-		tex_num_float_images = TEX_EXTENDED_NUM_FLOAT_IMAGES;
-		tex_image_byte_start = TEX_EXTENDED_IMAGE_BYTE_START;
-	}
-	else if((info.type == DEVICE_CUDA || info.type == DEVICE_MULTI) && info.extended_images) {
-		tex_num_images = TEX_EXTENDED_NUM_IMAGES_GPU;
-	}
-	else if(info.pack_images) {
-		tex_num_images = TEX_PACKED_NUM_IMAGES;
-	}
-}
-
 bool ImageManager::set_animation_frame_update(int frame)
 {
 	if(frame != animation_frame) {
@@ -267,9 +282,9 @@ int ImageManager::add_image(const string& filename,
 
 		if(slot == images.size()) {
 			/* max images limit reached */
-			if(images.size() == tex_num_images) {
+			if(images.size() == tex_num_byte_images) {
 				printf("ImageManager::add_image: byte image limit reached %d, skipping '%s'\n",
-				       tex_num_images, filename.c_str());
+				       tex_num_byte_images, filename.c_str());
 				return -1;
 			}
 

intern/cycles/render/image.h

@@ -32,7 +32,7 @@ class Progress;
 
 class ImageManager {
 public:
-	ImageManager();
+	ImageManager(const DeviceInfo& info);
 	~ImageManager();
 
 	int add_image(const string& filename,
@@ -62,7 +62,6 @@ public:
 
 	void set_osl_texture_system(void *texture_system);
 	void set_pack_images(bool pack_images_);
-	void set_extended_image_limits(const DeviceInfo& info);
 	bool set_animation_frame_update(int frame);
 
 	bool need_update;
@@ -86,7 +85,7 @@ public:
 	};
 
 private:
-	int tex_num_images;
+	int tex_num_byte_images;
 	int tex_num_float_images;
 	int tex_image_byte_start;
 	thread_mutex device_mutex;

intern/cycles/render/scene.cpp

@@ -54,7 +54,7 @@ Scene::Scene(const SceneParams& params_, const DeviceInfo& device_info_)
 	mesh_manager = new MeshManager();
 	object_manager = new ObjectManager();
 	integrator = new Integrator();
-	image_manager = new ImageManager();
+	image_manager = new ImageManager(device_info_);
 	particle_system_manager = new ParticleSystemManager();
 	curve_system_manager = new CurveSystemManager();
 	bake_manager = new BakeManager();
@@ -64,9 +64,6 @@ Scene::Scene(const SceneParams& params_, const DeviceInfo& device_info_)
 		shader_manager = ShaderManager::create(this, params.shadingsystem);
 	else
 		shader_manager = ShaderManager::create(this, SHADINGSYSTEM_SVM);
-
-	/* Extended image limits for CPU and GPUs */
-	image_manager->set_extended_image_limits(device_info_);
 }
 
 Scene::~Scene()

intern/cycles/render/scene.h

@@ -109,8 +109,8 @@ public:
 	device_vector<uint> sobol_directions;
 
 	/* cpu images */
-	device_vector<uchar4> tex_image[TEX_EXTENDED_NUM_IMAGES_CPU];
+	device_vector<uchar4> tex_image[TEX_NUM_BYTE_IMAGES_CPU];
-	device_vector<float4> tex_float_image[TEX_EXTENDED_NUM_FLOAT_IMAGES];
+	device_vector<float4> tex_float_image[TEX_NUM_FLOAT_IMAGES_CPU];
 
 	/* opencl images */
 	device_vector<uchar4> tex_image_packed;

intern/cycles/util/util_texture.h

@@ -21,26 +21,26 @@ CCL_NAMESPACE_BEGIN
 
 /* Texture limits on various devices. */
 
-#define TEX_NUM_FLOAT_IMAGES	5
+/* CPU */
+#define TEX_NUM_BYTE_IMAGES_CPU		1024
+#define TEX_NUM_FLOAT_IMAGES_CPU	1024
+#define TEX_IMAGE_BYTE_START_CPU	TEX_NUM_FLOAT_IMAGES_CPU
 
-/* generic */
+/* CUDA (Fermi) */
-#define TEX_NUM_IMAGES			88
+#define TEX_NUM_BYTE_IMAGES_CUDA	88
-#define TEX_IMAGE_BYTE_START	TEX_NUM_FLOAT_IMAGES
+#define TEX_NUM_FLOAT_IMAGES_CUDA	5
+#define TEX_IMAGE_BYTE_START_CUDA	TEX_NUM_FLOAT_IMAGES_CUDA
 
-/* extended gpu */
+/* CUDA (KEPLER and above) */
-#define TEX_EXTENDED_NUM_IMAGES_GPU		145
+#define TEX_NUM_BYTE_IMAGES_CUDA_KEPLER		145
+#define TEX_NUM_FLOAT_IMAGES_CUDA_KEPLER	5
+#define TEX_IMAGE_BYTE_START_CUDA_KELPER	TEX_NUM_FLOAT_IMAGES_CUDA_KEPLER
 
-/* extended cpu */
+/* OpenCL */
-#define TEX_EXTENDED_NUM_FLOAT_IMAGES	1024
+#define TEX_NUM_BYTE_IMAGES_OPENCL		1024
-#define TEX_EXTENDED_NUM_IMAGES_CPU		1024
+#define TEX_NUM_FLOAT_IMAGES_OPENCL		0
-#define TEX_EXTENDED_IMAGE_BYTE_START	TEX_EXTENDED_NUM_FLOAT_IMAGES
+#define TEX_IMAGE_BYTE_START_OPENCL		TEX_NUM_FLOAT_IMAGES_OPENCL
 
-/* Limitations for packed images.
- *
- * Technically number of textures is unlimited, but it should in
- * fact be in sync with CPU limitations.
- */
-#define TEX_PACKED_NUM_IMAGES			1024
 
 /* Color to use when textures are not found. */
 #define TEX_IMAGE_MISSING_R 1